GPT Image 2.5 API pricing calculator
Verified against OpenAI documentation on 2026-09-09
OpenAI bills GPT Image 2.5 by token, so there is no official price per image. This page puts the published token rates next to per-image figures from a provider that resells the model at a flat rate, and lets you work out a monthly number from both.
- Per image
- $0.0361
- Per request
- $0.0365
- Per month
- $36.54
Output $0.0361 plus input $0.00042 per request. Per-image figures are fal.ai’s, read on 2026-09-09. Input cost assumes about a thousand tokens per reference image, which OpenAI does not publish, so treat it as an estimate.
OpenAI token rates
Identical for gpt-image-2.5-flare and gpt-image-2.5-sunburst, and unchanged from GPT Image 2.
| Token type | Per 1M | Cached |
|---|---|---|
| Text input | $5.00 | $1.25 |
| Image input | $8.00 | $2.00 |
| Image output | $30.00 | n/a |
Per-image reference prices
fal.ai’s published prices for Sunburst, read on 2026-09-09. Longer prompts push these up. OpenAI publishes no equivalent table.
| Size | low | medium | high | xhigh | max |
|---|---|---|---|---|---|
| 1K · 1024 × 768 | $0.0040 | $0.0090 | $0.0361 | $0.0642 | $0.1444 |
| HD · 1920 × 1080 | $0.0044 | $0.0103 | $0.0396 | $0.0704 | $0.1584 |
| 4K · 3840 × 2160 | $0.0111 | $0.0260 | $0.1001 | $0.1779 | $0.4003 |
What $100 buys
Images at 1024 by 768, ignoring input cost. The spread across quality tiers is the reason to draft low and finish high.
| Quality | Images for $100 |
|---|---|
| low | 24,875 |
| medium | 11,074 |
| high | 2,768 |
| xhigh | 1,557 |
| max | 692 |
Frequently asked questions
Is Sunburst priced differently from Flare?
No. Both models bill at the same token rates: $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. Sunburst costs you time rather than money. Choosing it is a latency decision, not a budget one.
Are reference images charged?
Yes, as image input tokens at $8 per million. OpenAI does not publish a token count per reference image, so the calculator assumes about a thousand tokens each and labels the result an estimate. Measure your own usage response before you build a budget on it.
Does prompt length change the price?
Yes. Text input bills at $5 per million tokens, and providers note that longer prompts and requests that pull in more world knowledge cost more. The effect is small next to output cost at low quality, and smaller still at high quality, but it is real for long structured prompts.
What does cached input mean?
Repeated input can be served from cache at a lower rate: $1.25 per million text tokens instead of $5, and $2 per million image tokens instead of $8. It applies to input you send again, not to output. The calculator uses uncached rates, so treat its input figure as the ceiling.
Is there a free tier in the API?
No. API usage is billed from the first request. What is free is inside ChatGPT, where Images 2.5 went out to every tier on 8 September 2026. If you want to try the model without wiring up an API key, images25 gives new accounts ten image credits.
Where do the per-image numbers come from?
From fal.ai, read on 9 September 2026. OpenAI publishes token rates and not a price list, so a per-image figure has to come from a provider that resells the model at a flat rate. The token table on this page is OpenAI’s own.
How does the images25 credit price map to API cost?
One credit is one image at 1K; a 2K image is 2 credits and a 4K image is 3. Our upstream bills by resolution rather than quality tier, so the credits follow resolution too. The figures below are OpenAI’s own API prices, for comparison if you call the API directly.